Data center electric power and computing power collaborative optimization method and related device

Abstract

The invention relates to the technical field of electric power, in particular to a data center electric power and computing power collaborative optimization method and a related device, and the method comprises the steps: obtaining data processing pressure data of a data center and UPS operation data of the data center, and solving a pre-constructed electric power and computing power collaborative optimization problem model; and obtaining an optimal scheduling scheme with the goal of maximizing the weighted difference between the data processing amount and the energy consumption cost of the data center, and completing the collaborative optimization of the power and the computing power of the data center. According to the method, the dynamic service processing pressure of different servers of the data center and the influence of the UPS stability on the power and computing power collaborative optimization of the data center are fully considered, so that the data center meets the service requirements, and meanwhile, the energy consumption is reduced to the maximum extent. According to the method, the Lyapunov optimization theory and the genetic algorithm are combined for optimization, the optimization complexity is reduced, rapid convergence can be achieved, and the problems that an existing data center is low in electric power and computing power collaborative optimization accuracy, high in complexity and low in convergence speed are solved.

Description

Technical Field

[0001] The present invention relates to the field of power technology, and specifically provides a method and related device for collaborative optimization of power and computing power in a data center. Background Art

[0002] Data centers have typical high energy consumption characteristics and require a large amount of electricity to maintain the operation of their equipment. The stability and reliability of power supply are crucial for data centers. Computing power refers to the computing ability of a data center, which is the ability of the servers in the data center to output results after processing data. Computing power is a comprehensive indicator for measuring the computing ability of a data center, and the larger the value, the stronger the comprehensive computing ability. Achieving efficient coordination of power and computing power, especially introducing new energy power generation, on the one hand helps to reduce the operating cost of the data center, and on the other hand can effectively achieve the consumption of green electricity and maintain the stable operation of the power system.

[0003] However, the current research on power and computing power coordination for data centers is still in its infancy, mostly focusing on framework and strategic research. First of all, data centers aim at the efficient processing of business data, and the access of business data often has strong dynamics. For example, the data processing pressure is high during the day and low at night. The current research cannot fully combine the dynamic business processing pressure of different servers in the data center for power and computing power collaborative optimization. Secondly, data centers are usually equipped with Uninterruptible Power Supplies (UPS). Frequent overcharging and discharging processes will cause the batteries of the UPS to age and reduce their energy storage capacity, while the current research fails to consider the stable operation of the UPS. Finally, considering that the server clusters carried by data centers are very large, resulting in high complexity of power and computing power collaborative optimization. Therefore, there is an urgent need to develop a collaborative optimization method with high accuracy of optimization scheduling results and low complexity to optimize the energy consumption cost of data centers. Summary of the Invention

[0004] Aiming at the problems of low accuracy, high complexity, and slow convergence speed in the collaborative optimization of power and computing power in data centers in the prior art, the present invention provides a method and related device for collaborative optimization of power and computing power in a data center.

[0005] To achieve the above object, the present invention adopts the following technical solutions: The present invention provides a method for collaborative optimization of power and computing power in a data center, including: Obtaining data processing pressure data of the data center; Obtaining UPS operation data of the data center; Using the data center data processing pressure data and the data center UPS operation data, solve the pre-constructed power and computing power collaborative optimization problem model to obtain the optimal scheduling scheme with the maximization of the weighted difference between the data center data processing volume and the energy consumption cost, and complete the power and computing power collaborative optimization of the data center; Among them, the power and computing power collaborative optimization problem model is based on maximizing the weighted difference between the business data processing volume and the energy consumption cost on the premise of ensuring the stability of the server data queue and the UPS battery energy queue.

[0006] Optionally, the power and computing power collaborative optimization problem model is expressed as follows:

[0007] Among them, is the weight coefficient, which is used to adjust the order of magnitude of the business data volume and the electricity cost; is the real-time grid electricity price; is the server number of the data center; is the number of parallel processing units of each server; is the business data processing volume; is the electric energy purchased by the UPS from the power grid; is the business data arrival volume in the current time slot; is the minimum value of the business data arrival volume in the current time slot; is the maximum value of the business data arrival volume in the current time slot; is the computing resource provided by the corresponding server for the corresponding business processing unit in the current time slot; is the maximum available computing resource of the corresponding server; is the electric energy obtained by the UPS from the self-owned photovoltaic power station; is the maximum electric energy that the UPS can obtain from the self-owned photovoltaic power station; is the maximum electric energy that the UPS can purchase from the power grid; represents the mathematical expectation function; is the data backlog in the business processing queue; is the number of equal-length time slots for the overall optimization time division, and the duration of each time slot is ; is the stored energy in the UPS in the current time slot; is the safety energy storage capacity ratio, ; is the maximum energy storage capacity of the UPS.

[0008] Optionally, the process method for solving the pre-constructed power and computing power collaborative optimization problem model is: Based on the Lyapunov optimization theory, the power and computing power collaborative optimization problem model is transformed into a power and computing power collaborative optimization problem model that can be solved only relying on the current information; Based on the isolated genetic algorithm, solve the power and computing power collaborative optimization problem model that can be solved only relying on the current information, and obtain the optimal scheduling scheme with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost of the data center, so as to complete the power and computing power collaborative optimization of the data center.

[0009] Optionally, the power and computing power collaborative optimization problem model that can be solved only relying on the current information is expressed as:

[0010] Wherein, is the Lyapunov drift plus penalty function; is the term independent of the optimization variable.

[0011] Optionally, the method of transforming the power and computing power collaborative optimization problem model into a power and computing power collaborative optimization problem model that can be solved only relying on the current information based on the Lyapunov optimization theory is: Define the Lyapunov vector, and obtain the Lyapunov function related to the service processing queue and the energy queue; According to the Lyapunov function related to the service processing queue and the energy queue, define the Lyapunov drift function between time slots, and define the Lyapunov drift plus penalty function with the goal of maximizing the weighted difference between the service data processing volume and the energy consumption cost; According to the Lyapunov drift plus penalty function, transform the power and computing power collaborative optimization problem model into a power and computing power collaborative optimization problem model that can be solved only relying on the current information.

[0012] Optionally, the method of solving the power and computing power collaborative optimization problem model that can be solved only relying on the current information based on the isolated genetic algorithm, and obtaining the optimal scheduling scheme with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost of the data center, so as to complete the power and computing power collaborative optimization of the data center is: Define the eigenvalue vector of the service processing queue, and use the Euclidean distance to represent the similarity between two service processing queues; According to the similarity between two service processing queues, define the mean value of the similarities of all service processing queues; According to the mean value of the similarities of all service processing queues, obtain the dispersion degree of the entire server cluster; According to the mean value of the similarities of all service processing queues and the dispersion degree of the entire server cluster, divide all servers into several server subgroups; Taking the optimization variables of the power and computing power collaborative optimization problem model that only relies on current information for solution as the chromosome encoding, optimize the server subgroup to obtain the optimal solution for the computing resource allocation and power purchase quantity optimization of the server sub-cluster; According to the optimal solutions of the computing resource allocation and power purchase quantity optimization of all server sub-clusters, obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost of the data center on the premise of ensuring the stability of the service processing queue and the energy queue, and complete the power and computing power collaborative optimization of the data center.

[0013] The present invention provides a power and computing power collaborative optimization control system for a data center, including: A data processing pressure sensing unit: used to obtain the data processing pressure data of the data center; A UPS operation sensing unit: used to obtain the UPS operation data of the data center; An algorithm execution unit: used to utilize the data processing pressure data of the data center and the UPS operation data of the data center to solve the pre-constructed power and computing power collaborative optimization problem model, obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost of the data center, and complete the power and computing power collaborative optimization of the data center.

[0014] Optionally, the power and computing power collaborative optimization problem model is expressed as follows:

[0015] Among them, is the weight coefficient, used to adjust the order of magnitude of the service data volume and the electricity cost; is the real-time electricity price of the power grid; is the server number of the data center; is the number of parallel processing units of each server; is the service data processing volume; is the electric energy purchased by the UPS from the power grid; is the service data arrival volume in the current time slot; is the minimum value of the service data arrival volume in the current time slot; is the maximum value of the service data arrival volume in the current time slot; is the computing resource provided by the corresponding server for the corresponding service processing unit in the current time slot; is the maximum available computing resource of the corresponding server; is the electric energy obtained by the UPS from the self-owned photovoltaic power station; is the maximum electric energy that the UPS can obtain from the self-owned photovoltaic power station; is the maximum electric energy that the UPS can purchase from the power grid; represents the mathematical expectation function; Data backlog in the business processing queue; is the number of equal-length time slots divided for the overall optimization time, and the duration of each time slot is ; is the stored energy in the UPS during the current time slot; is the safety energy storage capacity ratio, ; is the maximum energy storage capacity of the UPS.

[0016] The present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0017] The present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for collaborative optimization of power and computing power in a data center. The method obtains data center data processing pressure data and data center UPS operation data; then uses the data center data processing pressure data and data center UPS operation data to solve a pre-constructed collaborative optimization problem model of power and computing power, and obtains an optimal scheduling plan with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost in the data center, realizing the collaborative optimization of power and computing power in the data center. Among them, the collaborative optimization problem model of power and computing power takes the weighted difference between the business data processing volume and the energy consumption cost as the optimization goal on the premise of ensuring the stability of the server data queue and the UPS battery energy queue. This method fully considers the stability of the server data queue and the UPS battery energy queue, ensures that the data center can maintain a stable operating state when facing various load changes, improves the accuracy of collaborative optimization, avoids system crashes or performance degradation caused by excessive or too little load, significantly reduces the operating cost of the data center, helps to reduce the carbon emissions of the data center, and conforms to the current global pursuit of green, low-carbon, and sustainable development; for the data center industry, this optimization method not only helps to improve economic benefits but also plays a positive role in environmental protection.

[0019] Regarding the power and computing power collaborative optimization problem model, in terms of computing power optimization, the model takes into account the amount of operation data in the data center at different time periods, and the scheduling scheme can be flexibly adjusted according to real-time or predicted data volume changes, which helps to ensure that the data center can efficiently process a large amount of data during peak hours, while reducing computing power consumption during off-peak hours, achieving reasonable allocation and utilization of resources; to avoid the overflow phenomenon caused by continuous data backlog, a long-term queue stability constraint is set on the server, which can ensure the smooth processing of data on the server and avoid the system overflow caused by continuous data backlog, which is crucial for maintaining the stability and reliability of the data center. In terms of power optimization, considering the computing energy consumption and cooling energy consumption of business data processing, a dynamic UPS energy queue model is established, and the photovoltaic power consumption and peak-valley electricity price are taken into account. To avoid the frequent overcharge and over-discharge processes from affecting the battery life, a long-term queue stability constraint of the energy queue at the UPS is set to ensure that it can reasonably allocate and store energy at different time periods. This helps to improve the energy efficiency of the UPS, extend its service life, and reduce maintenance costs, enabling the data center to make full use of renewable energy while meeting business needs, reducing dependence on the power grid, and storing energy or processing more data during off-peak electricity price periods, thereby reducing the overall electricity cost and minimizing energy consumption.

[0020] The long-term queue stability constraints related to the data processing queue and the energy queue make it difficult to directly solve the power and computing power collaborative optimization problem. The present invention directly introduces the Lyapunov optimization theory, sets the queue drift related to the long-term queue stability constraint, and combines the optimization objective of the original optimization problem. By minimizing the drift plus penalty, it is possible to ensure the satisfaction of the long-term queue stability constraint as much as possible while optimizing the utility. Moreover, the transformed optimization problem does not require any historical experience information and future prior information. Combining with the proposed power and computing power collaborative optimization system based on the centralized control architecture, it can perform online optimization only according to the information perceived in the current time slot. For the transformed data center power and computing power collaborative optimization problem, due to the coupling between optimization variables, it is a non-convex optimization problem that is difficult to directly solve. Therefore, a genetic algorithm is introduced to solve it. However, since the number of servers in the data center is often very large, directly applying the genetic algorithm to solve it has high complexity and poor convergence. Therefore, a power and computing power collaborative optimization method based on the isolated genetic algorithm is proposed. The concept of similarity is introduced to divide all servers in the data center into multiple small clusters, and the transmission of the optimization results is realized through chromosome replication, thereby effectively reducing the time complexity of the genetic algorithm execution. The chromosome replication between similar clusters can also greatly improve the convergence speed of subsequent optimization, solving the problems of high complexity and slow convergence speed in the power and computing power collaborative optimization of the data center in the prior art.

[0021] The present invention provides a data center power and computing power collaborative optimization control system. Through the highly integrated data processing pressure sensing unit, UPS operation sensing unit, and algorithm execution unit, it realizes the acquisition of data center data processing pressure data and data center UPS operation data, and uses the data center data processing pressure data and data center UPS operation data to solve the pre-constructed power and computing power collaborative optimization problem model, obtaining an optimal scheduling scheme with the goal of maximizing the weighted difference between the data processing volume and energy consumption cost of the data center. This system can real-time sense the operating state of the data center and intelligently adjust the allocation of power and computing power resources according to these state information, so as to achieve the purpose of reducing the energy consumption cost of the data center and improving the resource utilization efficiency. The system structure is simple, with fast real-time sensing and response speed, intelligent optimization scheduling, good cost-effectiveness, high resource utilization efficiency, good stability and reliability, and can effectively promote the data center to develop towards a more efficient, green, and sustainable direction.

[0022] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The characteristic is that when the processor executes the computer program, it realizes the steps of the above method. The device structure is simple, the transformation cost is low, and the resource occupancy is small.

[0023] A computer-readable storage medium stores a computer program. The characteristic is that when the computer program is executed by a processor, it realizes the steps of the above method. This storage medium has good portability and strong versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a data center model diagram of the present invention.

[0025] Figure 2 It is a schematic flow diagram of a data center power and computing power collaborative optimization method of the present invention.

[0026] Figure 3 It is a simplified flow diagram of a method for solving the Lyapunov power and computing power collaborative optimization problem model based on the isolated genetic algorithm in a data center power and computing power collaborative optimization method of the present invention, obtaining an optimal scheduling scheme with the goal of maximizing the weighted difference between the data processing volume and energy consumption cost of the data center.

[0027] Figure 4 It is a structure diagram of a data center power and computing power collaborative optimization system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0028] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0030] The present invention will be further described in detail below in conjunction with specific embodiments, which are explanations rather than limitations of the present invention.

[0031] See Figure 1 , which is a data center model diagram of the present invention. In the figure, data transmission devices such as switches are responsible for accessing service data into the server, the server is responsible for the analysis and processing of service data, and the uninterruptible power supply UPS is responsible for supplying power to the data center. Its power source includes two parts. One part is a new energy power generation system built by the data center or nearby, and the other part is the power purchased from the power grid enterprise. The UPS is equipped with a storage battery for energy storage to ensure that the data center can still be powered for a period of time when the external power is interrupted. Devices such as switches and servers will generate a large amount of heat during the transmission and processing of service data. Therefore, the data center is equipped with a large central air conditioner for heat dissipation and temperature reduction. For the above-mentioned data center, the existing power and computing power coordination methods of the data center cannot fully combine the dynamic business processing pressure of different servers in the data center for power and computing power coordination optimization, nor can they consider the stable operation of the UPS, resulting in problems such as low accuracy, high complexity, and slow convergence speed in the power and computing power coordination optimization of the data center.

[0032] See Figure 2 , in view of the above problems, the present invention discloses a method for optimizing the coordination of power and computing power in a data center, including: S1: Obtain the data processing pressure data of the data center, specifically: Data center computer rooms generally contain a very large number of matrix-style server arrays. Each server is connected to data transmission devices such as switches, and a huge amount of business data is accessed at each moment. The servers in the data center are represented by Each server has parallel processing units, represented by For each processing unit, the evolution of the time slots with backlogged business processing queues can be represented as:

[0033] Where, is the data backlog in the business processing queue; is the amount of business data arriving in the current time slot, satisfying , is the minimum value of the amount of business data arriving in the current time slot, is the maximum value of the amount of business data arriving in the current time slot, and the threshold range is related to the data access capabilities of data transmission devices such as switches; is the amount of business data processed, specifically represented as:

[0034] Where, is the computing resource provided by the server for this business processing unit in the current time slot, in GHz, and satisfies , is the maximum available computing resource of the corresponding server, related to the server model, server usage years, etc.; is the complexity of business data processing, in GHz / Gbits, that is, the average computing resource required to process one Gbit.

[0035] The energy consumption that can be optimized in the present invention includes the computing energy consumption required for the server to process business data and the energy consumption required for cooling after the server generates heat; Among them, the computing energy consumption required for the server to process business data is represented as :

[0036] Where, is the CPU energy consumption coefficient, mainly related to the hardware circuit conditions of the server; The energy consumption required for cooling after the server generates heat is represented as :

[0037] Where, is the heat energy consumption coefficient of the server, with the unit of J / Gbits, that is, the air-conditioning energy consumption required to process unit Gbits of service data on average. Its value is affected by many aspects, such as external factors like the geographical location, natural environment, and seasonal weather of the data center, and is also related to internal factors such as the building conditions of the data center and the degree of server aging. In actual optimization, it should be set in combination with the historical data of the data center itself.

[0038] Then, in one time slot, the total optimizable energy consumption of the data center is :

[0039] In the process of power-computing collaborative optimization, due to the limited storage capacity of the server itself, it should be ensured as much as possible that service data can be processed in a timely manner to avoid overflow caused by continuous backlog. Therefore, a long-term queue stability constraint on the server is set, that is:

[0040] Among them, is the mathematical expectation function.

[0041] S2: Obtain the UPS operation data of the data center, specifically: UPS is a special power supply containing an energy storage device. The part considered for optimization only includes the battery pack that supplies energy to the server and the air conditioner. The energy stored in the battery pack in the current time slot is The dynamic evolution of the energy queue between time slots in the relevant battery can be expressed as:

[0042] Among them, is the electric energy obtained by the UPS from the self-built photovoltaic power station. Affected by the output of the photovoltaic power at different times and line losses, its magnitude should not exceed a threshold, that is , is the maximum electric energy that the UPS can obtain from the self-built photovoltaic power station; is the electric energy purchased by the UPS from the power grid. Affected by the strength of the existing distribution network, the electric energy that can be obtained per unit time is also limited, that is , represents the maximum energy that can be purchased from the power grid.

[0043] In addition, the electric energy obtained from the self-built photovoltaic power station does not require electricity charges, and the electric energy purchased from the power grid implements peak-valley electricity prices. The service life of the battery is limited. In order to extend the service life, frequent overcharging and discharging processes should be avoided, and the battery power should be maintained within a stable range. Therefore, a long-term queue stability constraint on the energy queue at the UPS is set, that is:

[0044] The long-term saturation of the battery usually affects its service life, so a is set as the safety energy storage capacity ratio, and the specific value is related to factors such as the battery type.

[0045] S3: Use the data center data processing pressure data and the data center UPS operation data to solve the pre-constructed power and computing power collaborative optimization problem model, and obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data center data processing volume and the energy consumption cost, so as to complete the power and computing power collaborative optimization of the data center; Among them, the power and computing power collaborative optimization problem model is, on the premise of ensuring the stability of the server data queue and the UPS battery energy queue, with the weighted difference between the business data processing volume and the energy consumption cost as the optimization goal. From the perspective of the data center, the utility is set as the difference between the business data processing volume and the energy consumption cost. The purpose of the present invention is to calculate the power and computing power collaborative optimization problem model with the goal of maximizing the data center utility of the computing resources and the purchased power volume on the premise of ensuring the stability of the server data queue and the UPS battery energy queue as follows:

[0046] Among them, is the weight coefficient, which is used to adjust the order of magnitude of the business data volume and the electricity cost; is the real-time grid electricity price.

[0047] Then, in the method of solving the pre-constructed power and computing power collaborative optimization problem model to obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data center data processing volume and the energy consumption cost and completing the power and computing power collaborative optimization of the data center, due to the existence of the long-term constraints C5 and C6, the power and computing power collaborative optimization problem P1 is difficult to be directly solved, because it is difficult to accurately predict information such as the data arrival volume, photovoltaic power generation volume, and electricity price in the future time slots, and the cumulative effect between time slots is extremely likely to cause the performance of the data queue and the energy queue to deteriorate sharply. To solve this problem, the present invention introduces the Lyapunov optimization theory, transforms the long-term constraints into the optimization of the queue drift difference between time slots, and by minimizing the drift plus penalty, it is possible to ensure the satisfaction of C6 and C7 as much as possible while optimizing the utility. The specific method for solving the pre-constructed power and computing power collaborative optimization problem model is as follows: S3.1: Based on the Lyapunov optimization theory, transform the power and computing power collaborative optimization problem model into a power and computing power collaborative optimization problem model that can be solved only relying on the current information. This step is to transform the power and computing power collaborative optimization problem model with long-term queue stability constraints into a power and computing power collaborative optimization problem model that can be solved only relying on the current information based on the Lyapunov optimization theory. Specifically: S3.1.1: Define the Lyapunov vector and obtain the Lyapunov function related to the service processing queue and the energy queue; where the Lyapunov vector is :

[0048] The Lyapunov function related to the service processing queue and the energy queue is :

[0049] S3.1.2: According to the Lyapunov function related to the service processing queue and the energy queue, define the Lyapunov drift function between time slots, and define the Lyapunov drift plus penalty function with the goal of maximizing the weighted difference between the service data processing volume and the energy consumption cost The Lyapunov drift function between time slots is :

[0050] The Lyapunov drift plus penalty function :

[0051] where is the weight coefficient for balancing minimizing the drift and maximizing the optimization goal.

[0052] After derivation, it can be obtained that:

[0053] where is a term independent of the optimization variable:

[0054] S3.1.3: According to the Lyapunov drift plus penalty function, transform the power and computing power co-optimization problem model into a power and computing power co-optimization problem model that can be solved only relying on the current information. The power and computing power co-optimization problem model that can be solved only relying on the current information is:

[0055] S3.2: Based on the isolated genetic algorithm, solve the power and computing power co-optimization problem model that can be solved only relying on the current information, and obtain the optimal scheduling scheme with the goal of maximizing the weighted difference between the data center data processing volume and the energy consumption cost, and complete the power and computing power co-optimization of the data center. See Figure 3 , specifically: As can be seen from the above, there are still challenges in solving the optimization problem P2. First, there is a coupling between the two decision variables of the above optimization problem, resulting in a non-convex optimization problem that cannot be directly solved. Second, the number of servers in a data center is often extremely large, leading to an increase in the complexity of algorithm optimization and a slow convergence speed of the algorithm. To address this problem, the present invention uses an improved low-complexity genetic algorithm to solve the above optimization problem, which is named the isolation genetic algorithm. Among them, the genetic algorithm is a heuristic algorithm that searches for the optimal solution by simulating the biological evolution process. However, the large number of individual gene chromosomes will make the process of searching for the optimal solution more complex and extremely prone to falling into the sub-optimal solution situation. To achieve the dimensionality reduction solution of the optimization problem, the present invention introduces the concept of similarity to isolate each server. After isolation, the server clusters do not perform chromosome communication except for gene replication; S3.2.1: Define the eigenvalue vector of the service processing queue and use the Euclidean distance to represent the similarity between two service processing queues; among them, the eigenvalue vector of the service processing queue is defined as :

[0056] The similarity between two service processing queues represented by the Euclidean distance is :

[0057] Among them, are the eigenvalue vectors of the service processing queues defined for different processing units of different servers respectively; S3.2.2: According to the similarity between two service processing queues, define the mean value of the similarities of all service processing queues :

[0058] Among them, and represent different processing units respectively; S3.2.3: According to the mean value of the similarities of all service processing queues, obtain the dispersion degree of the entire server cluster :

[0059] Among them, is the mean value of the similarities of all servers; and They respectively represent different servers. Among them, the dispersion is used to guide the number of divisions of the server cluster. The greater the dispersion, the lower the similarity among all servers, and the smaller the number of server clusters should be to avoid affecting the optimization effect. On the contrary, the smaller the chromosome dispersion, the higher the similarity among all servers, and more small server clusters can be divided to improve the convergence speed.

[0060] S3.2.4: Divide all servers into several server subgroups according to the mean value of the similarity of all business processing queues and the dispersion of the entire server cluster. Specifically: Define the number of divisions of the server sub-clusters according to the dispersion of the entire server cluster as , randomly select servers and add them into different server sub-clusters in turn. Then, in the remaining servers, select the one with the lowest average similarity to the existing servers in the server sub-cluster and add it. Repeat this process until all servers are divided. At this time, it is necessary to ensure that the number of servers in all server sub-clusters is equal. If the total number cannot be divided evenly, discard the optimization of the extra part of the servers and ignore the impact of these small amounts of servers on the overall energy consumption of the data center.

[0061] S3.2.5: Optimize the server subgroups with the optimization variables of the power and computing power co-optimization problem model that only rely on the current information for solution as the chromosome encoding, and obtain the optimal solutions for the computing resource allocation and power purchase amount optimization of the server sub-clusters. Specifically: Randomly select a server subgroup for optimization. The genes on its chromosome encoding are the optimization variables of problem P2, that is, the computing resources allocated to all data queues in the server sub-cluster and the power purchase amount required to support the operation of the cluster. A set of genes that meet the constraint conditions constitutes a chromosome. The method is as follows: Initialization: Set the current generation number and the maximum number of iterations . Set the minimum population evolution rate. Under the premise of meeting the constraint conditions of optimization problem P2, randomly generate different chromosome encodings as individuals, individuals form the initial population. Note that at this time, constraints C3 and C4 are scaled down proportionally according to the number of divisions of the server sub-clusters.

[0062] Environmental fitness calculation: Calculate the environmental fitness of different individuals, and the environmental fitness is consistent with the optimization objective of optimization problem P2; Cross inheritance: Use the weight of the environmental fitness of different individuals in the sum of the environmental fitness of all individuals as the selection probability, and select a certain number of individuals for crossover, that is, randomly select a part of the chromosome segments at the same position between two individuals for exchange; Mutation: On the premise of meeting the constraint conditions, randomly modify some chromosome segments of some individuals in the population. After crossover inheritance and mutation, a new generation of population is formed. , and increment the corresponding generation number by one. Calculate the maximum environmental fitness of this generation, and divide the difference between the current maximum environmental fitness and the previous generation's maximum environmental fitness by the current maximum environmental fitness. The resulting value is the evolutionary rate of the current population. Termination: When the generation number reaches the preset maximum number of iterations, or when the evolutionary rate of the current population reaches the preset minimum population evolutionary rate, terminate the iteration and output the individual with the maximum environmental fitness at this time as the optimal solution for the calculation resource allocation and electricity purchase quantity optimization of the server sub-cluster. Chromosome replication: Copy the chromosome encoding of this individual to another unoptimized server sub-cluster as the initial population of this server cluster. Another server cluster executes genetic algorithm optimization and chromosome replication respectively according to the above method. Note that the initial population is no longer randomly generated until all server clusters have been optimized to obtain the optimal solutions for the calculation resource allocation and electricity purchase quantity optimization of all server sub-clusters. Through the isolation and division of server clusters and gene replication, the convergence speed of the genetic algorithm can be effectively improved, the overall optimization complexity can be reduced, and thus the efficiency of power and computing power co-optimization in the data center can be improved.

[0063] S3.2.6: According to the optimal solutions for the calculation resource allocation and electricity purchase quantity optimization of all server sub-clusters, obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data processing volume and energy consumption cost of the data center on the premise of ensuring the stability of the business processing queue and the energy queue, and complete the power and computing power co-optimization of the data center.

[0064] This method fully considers the impact of the dynamic business processing pressure of different servers in the data center and the stability of the UPS on the power and computing power co-optimization of the data center, establishes a dynamic data processing queue model, and conducts the co-optimization of power and computing power in the data center. It can not only achieve fast and accurate optimization, enable the data center to make full use of renewable energy while meeting business requirements, minimize energy consumption, contribute to reducing carbon emissions in the data center, and conform to the current global pursuit of green, low-carbon, and sustainable development.

[0065] See Figure 4 , the present invention also provides a power and computing power co-optimization control system for a data center, which is characterized by including: Data processing pressure perception unit: used to obtain data processing pressure data of the data center; UPS operation perception unit: used to obtain UPS operation data of the data center; Algorithm execution unit: It is used to utilize the data processing pressure data of the data center and the UPS operation data of the data center to solve the pre-constructed power and computing power collaborative optimization problem model, and obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost of the data center, so as to complete the power and computing power collaborative optimization of the data center.

[0066] Among them, the data processing pressure sensing unit is responsible for collecting the business queue information in the server in the current time slot, specifically including the business data access volume, business data backlog, server computing power, etc. The information sensed by the UPS operation sensing unit includes new energy power generation, battery power, etc. The algorithm execution unit executes the power and computing power collaborative optimization algorithm based on the sensed data processing pressure and UPS operation conditions, and issues decisions to the data center to guide specific data processing.

[0067] Through the highly integrated data processing pressure sensing unit, UPS operation sensing unit and algorithm execution unit, this system realizes the process of obtaining the data processing pressure data of the data center and the UPS operation data of the data center, and using the data processing pressure data of the data center and the UPS operation data of the data center to solve the pre-constructed power and computing power collaborative optimization problem model, and obtaining the optimal scheduling plan with the goal of maximizing the weighted difference between the data processing volume and the energy consumption cost of the data center. This system can real-time sense the operation state of the data center and intelligently adjust the allocation of power and computing power resources according to these state information, so as to achieve the purpose of reducing the energy consumption cost of the data center and improving the resource utilization efficiency. The system structure is simple, with fast real-time sensing and response speed, intelligent optimization scheduling, good cost-effectiveness, high resource utilization efficiency, good stability and reliability, and can effectively promote the data center to develop in a more efficient, green and sustainable direction.

[0068] The present invention provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in each of the above method embodiments. Or, when the processor executes the computer program, it implements the functions of each module / unit in each of the above device embodiments.

[0069] The computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention.

[0070] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory.

[0071] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0072] The memory can be used to store the computer program and / or modules. The processor realizes various functions of the terminal device by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory.

[0073] If the modules / units integrated in the terminal device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-described various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0074] The above are only the preferred embodiments of the present invention, and are not used to limit the technical solutions of the present invention. Those skilled in the art should understand that, without departing from the spirit and principle of the present invention, the technical solutions can be subject to several simple modifications and substitutions, and these modifications and substitutions also fall within the protection scope covered by the claims.

Claims

1. A method for collaborative optimization of power and computing power in a data center, characterized in that: include: Obtain data processing pressure data of data centers; Obtain data center UPS operation data; Using the data center data processing pressure data and the data center UPS operation data, the pre-built power and computing power collaborative optimization problem model is solved to obtain the optimal scheduling solution with the goal of maximizing the weighted difference between the data center data processing volume and energy cost, thus completing the data center power and computing power collaborative optimization; Among them, the electricity and computing power collaborative optimization problem model is based on the premise of ensuring the stability of the server data queue and the UPS battery energy queue, with the optimization goal of maximizing the weighted difference between the business data processing volume and the energy consumption cost.

2. The data center power and computing power collaborative optimization method according to claim 1 is characterized in that: The power and computing power collaborative optimization problem model is expressed as follows: in, is the weight coefficient, which is used to adjust the magnitude of the business data volume and electricity cost; Real-time electricity price for the power grid; Number the servers in the data center; is the number of parallel processing units on each server; The amount of business data processed; The power purchased by UPS from the grid; is the amount of service data arriving in the current time slot; is the minimum value of the service data arrival amount in the current time slot; The maximum value of the service data arriving in the current time slot; The computing resources provided by the server corresponding to the current time slot for the corresponding service processing unit; is the maximum available computing resource of the corresponding server; The power obtained by UPS from its own photovoltaic power station; The maximum power that the UPS can obtain from the self-provided photovoltaic power station; The maximum amount of power that the UPS can purchase from the grid; represents the mathematical expectation function; Processing queue data backlog for business; The number of time slots of equal length divided into the overall optimization time, and the maintenance time of each time slot is ; Store energy in the UPS for the current time slot; For the safe energy storage capacity ratio, ; It is the maximum energy storage capacity of UPS.

3. The data center power and computing power collaborative optimization method according to claim 2 is characterized in that: The method for solving the pre-built power and computing power collaborative optimization problem model is: Based on Lyapunov optimization theory, the power and computing power collaborative optimization problem model is transformed into a power and computing power collaborative optimization problem model that only relies on current information for solution; Based on the isolated genetic algorithm, the power and computing power collaborative optimization problem model that only relies on current information is solved, and the optimal scheduling plan with the goal of maximizing the weighted difference between the data center's data processing volume and energy cost is obtained, thus completing the collaborative optimization of power and computing power in the data center.

4. The data center power and computing power collaborative optimization method according to claim 3 is characterized in that: The power and computing power collaborative optimization problem model that relies only on current information to solve is expressed as: in, Add penalty function for Lyapunov drift; is an item that is independent of the optimization variable.

5. The data center power and computing power collaborative optimization method according to claim 3 is characterized in that: The method of converting the power and computing power collaborative optimization problem model into a power and computing power collaborative optimization problem model that relies only on current information for solution based on the Lyapunov optimization theory is as follows: Define Lyapunov vectors and obtain Lyapunov functions related to business processing queues and energy queues; According to the Lyapunov function related to the service processing queue and the energy queue, the Lyapunov drift function between time slots is defined, and the Lyapunov drift plus penalty function is defined with the goal of maximizing the weighted difference between the service data processing volume and the energy consumption cost; According to the Lyapunov drift plus penalty function, the power and computing power collaborative optimization problem model is transformed into a power and computing power collaborative optimization problem model that only relies on current information for solution.

6. The data center power and computing power collaborative optimization method according to claim 3 is characterized in that: Based on the isolation genetic algorithm, the power and computing power collaborative optimization problem model that relies only on current information is solved to obtain the optimal scheduling solution with the goal of maximizing the weighted difference between the data center data processing volume and the energy cost. The method for completing the power and computing power collaborative optimization of the data center is: Define the eigenvalue vector of the business processing queue, and use the Euclidean distance to represent the similarity between two business processing queues; According to the similarity between two business processing queues, the mean value of the similarity of all business processing queues is defined; Obtain the dispersion of the entire server cluster based on the mean of the similarities of all business processing queues; According to the mean of similarity of all business processing queues and the dispersion of the entire server cluster, all servers are divided into several server subgroups; The optimization variables of the power and computing power collaborative optimization problem model that relies only on current information are used as chromosome codes to optimize the server sub-cluster and obtain the optimal solution for the server sub-cluster computing resource allocation and power purchase optimization. Based on the optimal solution for resource allocation and power purchase optimization of all server sub-clusters, the optimal scheduling plan is obtained with the goal of maximizing the weighted difference between data center data processing volume and energy consumption cost while ensuring the stability of business processing queues and energy queues, thus completing the coordinated optimization of data center electricity and computing power.

7. A data center power and computing power collaborative optimization control system, characterized in that: include: Data processing pressure sensing unit: used to obtain data processing pressure data of the data center; UPS operation sensing unit: used to obtain UPS operation data in the data center; Algorithm execution unit: It is used to solve the pre-built power and computing power collaborative optimization problem model by using the data center data processing pressure data and the data center UPS operation data, and obtain the optimal scheduling plan with the goal of maximizing the weighted difference between the data center data processing volume and the energy cost, so as to complete the data center power and computing power collaborative optimization.

8. The data center power and computing power collaborative optimization control system according to claim 7 is characterized in that: The power and computing power collaborative optimization problem model is expressed as follows: in, is the weight coefficient, which is used to adjust the magnitude of the business data volume and electricity cost; Real-time electricity price for the power grid; Number the servers in the data center; is the number of parallel processing units on each server; The amount of business data processed; The power purchased by UPS from the grid; is the amount of service data arriving in the current time slot; is the minimum value of the service data arrival amount in the current time slot; The maximum value of the service data arriving in the current time slot; The computing resources provided by the server corresponding to the current time slot for the corresponding service processing unit; is the maximum available computing resource of the corresponding server; The power obtained by UPS from its own photovoltaic power station; The maximum power that the UPS can obtain from the self-provided photovoltaic power station; The maximum amount of power that the UPS can purchase from the grid; represents the mathematical expectation function; Processing queue data backlog for business; The number of time slots of equal length divided into the overall optimization time, and the maintenance time of each time slot is ; Store energy in the UPS for the current time slot; For the safe energy storage capacity ratio, ; It is the maximum energy storage capacity of UPS.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images